HURRAY: Nvidia 600 series not just Fermi!! (Kepler)

The 3720qm is a fair bit faster than a stock 2960xm, not sure where you came up with that estimation Cloud..

 

WoW, 680M is really impressive!
Gets 1000+ points more than 6990M in 3Dmark11, Amazing

Sadly, I think it will be pricey! :/

 

Clock speed doesn't equal to performance, cache-wise it is 6mb L3 against 8mb L3, I didn't see any benches yet, but I am sure it won't be able to perform 2960xm level (even not 2920xm level). Also xm cpu has unlocked multipliers, you can easily make it +4GHz turbo

EDIT: just checked, 2960xm is better in every spec, where did you read it is faster?

 

non-xm cpus throttle a ton. in the grand scheme of things, the 2920/2960 are vastly superior CPUs to 27xx series and thus any non XM ivy mobile.

 

"In Crysis Warhead (GPU-limited), we were able to achieve with this combination in 720p with no anti-aliasing and anisotropic filtering an average fps rate of 29."
(Google translate)
Is this for real ?
Kepler-Notebook mit Nvidia GT 650M: Erste Benchmarkwerte von PCGH

 

Google Translate

Macbook skimping on memory again.

 

wow, just wow.. ultrabooks running warhead tech is fast progressing

 

wait i might just be an idiot but where does it say ultrabook lol

 

well, not exactly this, but acer timeline ultra (an ultrabook) has 640m, which is very similar to 650m, and if 650m is pulling off 29fps, I bet you can OC 640 to 25 , so there you go! ultrabook + crysis

 

well doesnt the acer already have heatig problems lol. and i thought the 640m is ddr3. Idk if we will be able to oc any of these cards up to the next level cause they put lower cards in for a reason: cooling. if the laptop could support the cooling of 650m speeds then there would be 650m in there

but if we are able to thats better for us

 

2860QM, 2960XM, 3720QM, pretty much all the same.

2860QM 2.5-3.6GHz
2960XM 2.7-3.7GHz
3720QM 2.6-3.6GHz

You get around 5% better speed with Ivy due to architectural improvement but I don`t see them perform much different.

 

I don't think 12k is a highscore if you know the GDDR3 version already scores 10k and is heavily bottlenecked by its 28.8GB/s memory bandwidth. The GDDR5 version should atleast give an 30 to 40% boost in reallife performance since it doubles the memory bandwidth.
But 2900 points for the CPU is really low (3dmark06). My i5 520M scores around that. So it was def not a 2760QM (which should also run at 2.6ghz and not 2.2ghz stock)

So the GPU is bottlenecked by CPU now instead of crappy GDDR3...

 

SO according to the OP; you can OC the Kepler based cards up to 20% based on stock software?? That would make the 660m capable of near 675m speeds maybe not at 1080p , but at 900p probably [considering the 128bit to 192bit difference? ]

 

I should remove that picture. I don`t think that software was included with GTX 680 and Kepler.
There is a guy somewhere in this forum that overclocked his GT 650M with stable temperatures and he got a decent boost in performance. It is somewhere in this thread. I`m too lazy to track it down

 

cloudfire
3610qm with gtx670m fermi
Google Translate

and 3612qm with gtx670m
Google Translate

Looks only 500 difference in 3dmark06 which is decent amount but most people will prefer to get the lower powered 3612qm. The gpu does better with the 3612qm strangely maybe less heat= more performance or maybe just luck with the benchmark.

 

Don`t look too much in to 3DMark06. Its difficult to comprehend the results. Look at cinebench or something like that instead. The 3610QM is clocked 200MHz higher than 3612QM. That is the difference. 3612QM is very interesting yes since its only 35W. I will await real reviews from Anandtech etc before making a decision.
Don`t think less heat means more performance with GTX 670M since its a Fermi. Unless its really hot. Prolly just a coincidence. I don`t know

This is probably why Nvidia haven`t announced any other Kepler chips than GK107 for mobile. And why we haven`t heard anything at all from AMD. And why GTX 680 is sold out almost everywhere. Also rumors are floating around on the internet that Nvidia is now using the damaged chips to make lower end chips that were originally intended for GK104 and such.
Good job guys. Not the first time you are screwing up things TSMC. Amateurs!

TSMC 28nm Capacity Issues Causing Nvidia Kepler GPU Launch Delays

 

Q3 2012? dkfhadskjfjlsakdfkjdsafaskjflask wow TSMC

 

Hey any idea if OC of the 650m will perform as 660m, or maybe a bit more, seems like 650 is the same card, with lower clocks.

 

If you get the GDDR5 version, yeah that should work. After all its the same chip they are using with 650M and 660M

 

I believe you may be misreading it. If you look at the source Cloudfire quoted, it says:

So everything should be 100% by at least July, and maybe sooner. Personally, I'd expect it to be June, and saying Q3 is just a convenient way of not getting expectations set too high.

 

dude, sorry but Q3 ends by September.. anyway I am sure we will see mobile chips by Fall 2012

 

more than likely, it will be in the market atleast by august maybe JUly, cause they want to capitalize on the Backto school sales

wouldnt the cooling matter also? the 650m probably was put in because thats how much the cooling could handle and not 660m speeds

 

It says ends by Q3, not by the end of Q3.
As PaKii94 says, it wouldn't make any sense for them to miss back-to-school sales, either.

 

dude, always aim for the end of a quarter when somebody says ready by this quarter, i.e. diablo 3 releasing Q2 = May 15th (still mid quarter, very optimistic, that is because they already had 4 years of development)

 

Not gonna fight about it; it will be what it will be.

Anyway, I'll just leave this here.

@Nvidia/AMD:

If you're havin' quarter problems I feel bad for you, son.
I got 99 problems, but a chip ain't one.

 

^^ nice one

I indeed hope for an asap release, but you know when it comes to these companies nothing is for sure at the best of times, with TSMC screwing things up, wouldn't count on it anyway we were supposed to get the highend first, then the mainstream, but with all this yield and production problems, now we can get the lowend first, then highend this was exactly the rumor last summer, when people started to hear how bad things are at TSMC, otherwise we were supposed to get 28nm in 2011, not 2012..

 

No wonder why Nvidia want to move to Intel factories and produce their 28nm chips there

 

Does Intel even have 28nm fabs? I think they skip all the half nodes. Shame all the other foundries can't keep pace with Intel, would be great to see 22nm GPUs.

 

Nvidia is dreaming. Even if Intel agrees, the fabs do either 32nm or 22nm wafers only. That requires a different design from 28nm.

 

Cloud we both have mediocre 720qm's and are looking at Ivy Bridge for things like pcsx2/dolphin emulation, and as such the chips performance is of paramount importance to us. Remember that Ivy bridge brings increased turbo performance for 2cores/3-4 cores, so the 3720qm will be 3.5Ghz(2c)/3.4Ghz(3-4c), which is the same as the 2960xm. Then factor in the IB performance increase at clock for clock, which ranges from 5%-15%, and you get the 3720qm being 5-15% better than a stock 2960xm. 8mb of l3 cache will not make a noticeable difference, and as such we have seen that the 2760qm is a superior option to the 2820qm due to being clocked 100mhz higher, not to mention being officially cheaper.

 

NVIDIA GK106 GPU Detailed

With the 130W power consumption it could really become a mobile GTX680M, sadly it has only 24 Rops instead of 32 and 192-bit bus instead of 256-bit.

 

yeah, some guy also corrected the wiki page with this information (24ROP and 192bit bus) a few days ago, I was like, what?

 

Real quick, Ive gotten a lot of information on the 650M card from the forums, but am stuck deciding between a Sager 6165 with the 650M GDDR5 and the 9130 with the 670M GDDR5. The price difference to the 9130 is about $150 give or take, but I keep getting the feeling that it's not worth investing in the 670M simply because of its older fermi design...

It seems silly, but its pretty obvious the 670M Fermi outperforms the 650M Kepler... so then... What exactly is the advantage of going with the Kepler, aside from sacraficing performance to save a little money?

 

Intel is just starting to offer their fabs on a trial basis. FPGA firm to use Intel's 22-nm fabrication technology - The Tech Report

Although as the article points out, it may be for their own benefit rather than exploring a new revenue stream. However, both AMD and Nvidia would love access to Intel's process lead, and they could steal a lot of revenue in a market they don't compete in (discrete GPUs). They could also make a lot on mobile chips, but it seems unlikely they would let ARM challenge their Trail mobile products.

 

dafuq? Why would it be advertised as a 2.6ghz processor if it has 3.4ghz 3-4 core speed?

 

That is pretty awesome. A slight downclocking and we got ourself what could be the next 100W GPU from Nvidia.

Nvidia is correcting the bandwidths with faster memory btw. So its not so bad, although they could have chosen not to cheap out

 

Still tho, the 660 seems to be slower than 7850 from that chart. And they are both at the same TDP?

Hmmm

 

I can see it now;


AMD: Hey! You guys are giving us the CRAP off your production line!
Intel: What makes you say that?
AMD: Look, this wafer has a Hair in it!
Intel: So.. Its not like its a pube or anything..
AMD: GRRRRRRR!!!

 

Hmmm, good catch. I don`t know if that chart is true or not, but who knows, AMD might have the upper hand this round. 7850 is a TDP of 130W so yeah, same TDP as GTX 660.

We must also not forget that 480M and 485M was both Fermi and both had a TDP of 100W yet 485M is miles ahead of 480M. Who knows, they might do the same "mistake" again lol

 

can agre compleatly had both in sli still have the later awsome card i think its a good ide to wait a bit if the gk106 is the new 680m becouse its prob a f up on nvidias part i could have supported a gk106 with 256 not the bs we are getting im dissapointed in nvidia it looks like they think we dont need performance hope clevotakes mathers in there own hands do something nice for us

 

Having the sweet spot chip at ~1/2 the high-end spec-wise is not great as it leaves too wide a gap between the two. The latest example I can remind of was the Radeon 5770 <=> 5850/5870. Though to be fair nvidia might not have planned that the GK104 would be to occupy the high-end for a prolonged period.

That's how turbo boost works. Since the clocks you're able to reach with it are not guaranteed (though stable at some point) and trigger under special conditions that depend upon the CPU use and thermals afaik, they'd rather advertise the base clocks. Current Sandy Bridge mobile chips are already flirting with 3.0ghz or even higher with turbo.

 

750mhz 256bit GDDR5 = 1000mhz 192bit GDDR5

They would have identical bandwidth.

 

how does this chart apply to the mobile chips?

 

Because the GTX 660 will most likely be used in the 680M, as I do believe the 580M was basically the 560Ti

 

True for both sides. The flagship mobile GPU tends to be based on the 3rd most powerful desktop card (560, 7870).

 

None, if you want perfomance get the 670m.. aslong as there isnt any high end kepler equivalent you shouldnt consider getting a midend kepler to replace the high end fermis..

 

Get a gtx660m. half power consumption. You have to be really dumb to get a 670m or 675m. With a 7970m or 7950m or gtx680m already benchmarked and very close to release date compared to the 1-2 year old design of the 670m or 675m cards.

 

Really? As recently as a couple months ago (when I first started researching laptops) people loved the performance of the GTX580m, just not the price. The GTX675m has the same performance but at a much cheaper price. Keep telling people they're dumb, kiddo. Just because you care about efficiency doesn't mean other people have the same mentality.

If there was a negative rep button, I'd push it so much you'd think we were in a late 80's Salt-n-Pepa video.

 

Now Kids.. Lets play nice..

 

Oh god... I would give my right arm for some legit "leaks" right now